Tagging Practices on Research

Oriented Social Bookmarking Sites

CAIS/ACSI 10-12 Mai 2007, Montréal, QC

Margaret E. I. KippFaculty of Information and Media Studies

University of Western Ontariomargaret.kipp@gmail.com

Background● My research examines:

– how people organise things on the web– how this compares to traditional library classification

techniques● Specific points of interest:

– structures and the creation of structures in classification systems

– relationship between personal information management and classification

Social Bookmarking and Tagging● Social Bookmarking:

– site for sharing bookmarks, articles, etc.– association of tags (keywords) with links– tags and articles are joined into networks of related

terms– users are encouraged to share bookmarks and tags

with others● Tagging:

– the act of associating a term with a link or article– labelling or classifying for personal use

Social Bookmarking Post

A post is a relationship between a user, an item and a set of tags.

Previous Studies● Study 1: Del.icio.us● study of Del.icio.us

tag usage on highly tagged sites

● examination of convergence of tag usage

● co-occurrence analysis for co-used tags

● Study 2: CiteULike● study of CiteULike tag

usage in comparison to author keywords and subject headings from on-line journal databases

● examine types of tags and more traditional index terms

Commonalities Between Studies● Study 3: Del.icio.us, CiteULike, Connotea● use of affective tags (e.g. cool, interesting) and

time and task related tags (e.g. @toread, todo) in both studies

● > 16% of tags in Del.icio.us study● average of 1-3 tags per article in CiteULike

study were not directly subject related● categories: time and task, affective, geographic,

methodology, emergent vocabulary, other

Motivations● Builds on study 2 of CiteULike

– Kipp (2006): users do use words from thesaurus as tags, but often use similar or related terms from other fields

● Examine use of indexing terms by users, authors and intermediaries

● Do they appear to provide a similar context?

Related Studies● Mathes (2004): suggested examination of user,

author and intermediary terms● Voorbij (1998), Ansari (2005): relatively high

degree of match between descriptors (intermediary terms) and title keywords

● Kipp (2006): found differences in term usage between users, authors and intermediaries

Research Question● To what extent do term usage patterns of user

tags, author keywords and intermediary descriptors suggest a similar context between users, authors and intermediaries?

Data Collection● Tag data collected from CiteULike● Journals used: Proteins and Journal of

Molecular Biology● Data collected included DOI or URL for

collection of keywords and descriptors● Author Keywords and Pubmed Descriptors from

journal sites and Pubmed respectively● 1083 articles (1588 posts) collected

Data Analysis● Informetric analysis using SQL (see Wolfram

2005)– standard informetric measures: frequency of

occurrence of unique tags● Thesaural analysis (see Voorbij 1998, Kipp

2006)– comparison of terms using Pubmed thesaurus

(range from SAME, SYN, NT, BT, RT, related and Not related)

Authors, Users and Journals● Authors:

– 80% of articles had between 2 and 5 authors– one article had 48 authors

● Users:– 239 unique users, 1588 posts– most prolific user had posted 94 posts– 42 users posted 10 or more articles

Tags, Keywords and Descriptors

● ratio of unique terms to total terms highest for author keywords

● supports findings from previous study in which author keywords were found to be more diverse than tags or descriptors

Tags Keywords DescriptorsUnique 1136 3181 2746Total 3788 4866 12473

Popular Tags, Keywords and Descriptors

Tags Frequencyprotein_structure 140no-tag 114protein 114structure 103docking 97

Author Keywords Frequencyprotein folding 58protein structure 49molecular dynamics 46protein structure prediction 38docking 31

Descriptors FrequencyModels, Molecular 649Protein Conformation 511Proteins 388Amino Acid Sequence 306Binding Sites 280

● 645 tags were used only once in the data set

● 2548 keywords were used only once

● 731 descriptors were used only once

● keywords are more diverse

Tags, Keywords and Descriptors by Article

● maximum number of tags per article was 29, minimum 1 and median 2– article with 29 tags was tagged by 14 users (most

users still use 1-3 tags to an article)● maximum number of tags per post was 15,

minimum 1 and median 2● maximum number of keywords per article in the

data set was 13, minimum 1, median 5● maximum number of descriptors per article was

36, minimum 2, median 11

User Vocabulary Length

User Max tag listlength

Min tag listlength

Number of articlesposted

3109 7 2 153063 6 1 734068 15 2 9

● user vocabulary length: the number of unique terms used by a user

● largest user vocabulary length was 62 (min. 1, median 2)

● most users use a small number of tags

Thesaural AnalysisTags Keywords Descriptors3d 16 S RNA Base Sequencealgorithms ribosome Computer Simulationprediction computer

modelingCross-Linking Reagents

rna distancegeometry

Escherichia coli

16s Models, Moleculardistance_geometry Molecular Sequence Databioinformatics Nucleic Acid Conformationstructure RNA, Ribosomal, 16Sstructure_prediction

● Article 788: Computer modeling 16 S ribosomal RNA● across all three sets of terms are variants on RNA and 16s● use of term bioinformatics versus computer

modelling/simulation

Discussion● results from the previous study (Kipp 2006)

using a smaller data set from library science are relevant to other fields and to larger data sets

● users use some terminology which is rare or completely absent from author keyword lists or descriptor lists (e.g. time and task tags)

● user terms often not part of formal thesaurus– 'protein' and 'structures' as separate tags were

linked in the thesaurus– abbreviations such as 'PDB' for 'Databases, Protein'

Discussion 2● tags: 'human', 'animal', and 'family-studies'

showed users tagging biology related articles are extremely interested in methodology and user groups associated with articles, this did not occur in the previous study

Margaret E. I. KippFaculty of Information and Media StudiesUniversity of Western Ontariomargaret.kipp@gmail.comhttp://publish.uwo.ca/~mkipp/

Thank you/Merci!

Questions?

CAIS/ACSI 10-12 Mai 2007, Montréal, QC